refactor(rag): 梳理文档入库与检索服务结构

tests/test_agent_core.py

@@ -1,5 +1,5 @@
 from agent_core.orchestrator import build_messages, run_agent
-from agent_core.rag.ingest import ingest_document
+from agent_core.rag.ingest import _split_text, ingest_document
 from agent_core.rag.retriever import retrieve
 
 
@@ -221,3 +221,30 @@ def test_run_agent_uses_retrieved_document_chunks(tmp_path):
 
     assert result.references[0]["source"] == "sop.md"
     assert "隔离现场" in result.references[0]["content"]
+
+
+def test_rag_split_text_keeps_overlap_and_non_empty_chunks():
+    chunks = _split_text("A" * 20, chunk_size=8, overlap=3)
+
+    assert chunks == ["AAAAAAAA", "AAAAAAAA", "AAAAAAAA", "AAAAA"]
+
+
+def test_retrieve_returns_empty_when_query_has_no_overlap(tmp_path):
+    store_path = tmp_path / "rag_store.json"
+    ingest_document(
+        scenario_id="knowledge_qa",
+        source_file="rules.md",
+        text="这里描述的是报销流程和审批链。",
+        collection="knowledge_qa",
+        store_path=store_path,
+    )
+
+    chunks = retrieve(
+        scenario_id="knowledge_qa",
+        query="设备点检",
+        collection="knowledge_qa",
+        top_k=3,
+        store_path=store_path,
+    )
+
+    assert chunks == []

tests/test_documents.py

@@ -3,7 +3,7 @@ from django.urls import reverse
 
 from apps.documents.forms import DocumentUploadForm
 from apps.documents.models import UploadedDocument
-from apps.documents.services import extract_text
+from apps.documents.services import extract_text, index_document
 
 
 def test_upload_txt_document_creates_uploaded_record(client, db):
@@ -128,3 +128,20 @@ def test_index_failure_message_is_visible_on_document_list(client, db, monkeypat
     assert response.status_code == 200
     assert "文档入库失败，请检查错误原因后重试" in content
     assert "模拟入库失败" in content
+
+
+def test_index_document_marks_failed_when_extracted_text_is_empty(db, monkeypatch):
+    document = UploadedDocument.objects.create(
+        scenario_id="knowledge_qa",
+        original_name="empty.md",
+        file_type="md",
+        size=0,
+        status="uploaded",
+    )
+
+    monkeypatch.setattr("apps.documents.services.extract_text", lambda target: "   ")
+
+    updated_document = index_document(document)
+
+    assert updated_document.status == UploadedDocument.STATUS_FAILED
+    assert "文档内容为空" in updated_document.error_message